Skip to content
Four-stage framework covering WisTMP dataset construction, model selection, fine-tuning, and local deployment.

Large Language Model-Assisted Preparation of Transportation Management Plans: A Case Study with WisDOT WisTMP System

Sheng, Z., Li, P., Huang, Z., Chen, Y.-J., Luo, Y., Wan, Z., Parker, S. T., Noyce, D. A., & Chen, S. (2026).

arXiv preprint, arXiv:2610.10650.

Summary

This study explores how locally deployed language models can help prepare transportation management plans for work zones. Historical WisDOT WisTMP documents are converted into question-answer training data, and several open-source models are fine-tuned to draft plan content. Fine-tuning improves text-generation metrics, but the models still tend to suggest too many strategies and struggle with project-specific explanations and cost estimates. Larger models provide limited additional gains, highlighting both the potential and remaining limitations of AI-assisted plan preparation.

Summary source

Four-stage framework covering WisTMP dataset construction, model selection, fine-tuning, and local deployment.
Figure 1. Overview of the LLM-assisted transportation management plan preparation framework (preprint version).